MCP server for Azure DevOps integration, enabling search and file reading capabilities across Azure DevOps repositories
This Azure DevOps MCP server has basic tool definitions but lacks several critical quality attributes. Both tools have names that follow verb_noun convention (search, read), which is correct. However, descriptions are generic and repetitive, parameter validation is missing, output schemas are undocumented, and error handling is minimal. The server exposes configuration via environment variables (good for secrets), but the tool interface itself does not match production standards for LLM interaction. Code inspection shows tools are explicitly registered in main.go with mcp.NewTool(), so definitions are directly visible (not inferred).
Read file content from Azure DevOps. The key to getting this to work well is asking for at least 5 results from the search tool, then asking specifically for code examples
Search for files in Azure DevOps repositories. The key to getting this to work well is asking for at least 5 results from the search tool, then asking specifically for code examples
Descriptions are repetitive and include LLM anti-patterns (example values). Both tools repeat identical instruction text: 'The key to getting this to work well is asking for at least 5 results from the search tool, then asking specifically for code examples.' This is an example value that LLMs will hallucinate into actual calls, violating pattern:tool-description. Descriptions should state WHAT the tool does, WHEN to use it, and any prerequisites, not how the LLM should prompt itself.
Output schemas are not documented. Neither tool specifies what fields are returned or what structure the LLM should expect. Code shows search returns []map[string]interface{} with fields like 'repository', 'path', 'fileName', 'project', but this is invisible in the tool definition. The read tool returns a string (file content), but this is undocumented. LLMs cannot plan downstream tool calls without knowing output structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 49 | - | v1 |
Parameter descriptions are minimal and lack formatting constraints. 'search' has params 'query' (description: 'Search query') and 'repo' (description: 'Optional repository name to search in'). These lack guidance on expected format, length, character restrictions, or examples of valid values.
No error handling guidance. Code shows error returns are raw fmt.Errorf() with messages like 'repository not found', 'error searching code', 'error getting file content'. These tell the LLM nothing about whether to retry, ask the user, or abort. Per pattern:recovery-guide, errors must guide the LLM to the next action.
Result limits not enforced or documented. The search tool sets Top: 1000 in the Azure API call, but the tool description does not state 'Results capped at 1000' or offer pagination parameters (offset, limit). Returning 1000 items could exhaust context windows. Per mxe:enforce-result-limits, cap results at 20-50 and offer pagination.
Tool names are generic. 'search' and 'read' are bare action verbs without a domain qualifier. Better names would be 'search_azure_devops_repositories' or 'search_code' and 'read_azure_devops_file' or 'read_code'. This becomes critical if agents have many tools, bare names like 'search' and 'read' conflict with other tool sets.
No chaining IDs in output. If search returns repository and path, the read tool needs both parameters (repository, path). Code shows this is the case, but the search result structure is undocumented, so the LLM cannot see that it can pass the response fields directly to read.
read tool requires both 'repository' and 'path', but search returns results with 'repository', 'path', 'fileName', 'project'. It's unclear which combination the read tool expects. Does it require the full path including branches? Code shows it calls GetItem with a path, but the tool description says nothing about expected path format (e.g., must include branch name, or relative to repo root).